drivers/block/virtio_blk.c

   1 //#define DEBUG
   2 #include <linux/spinlock.h>
   3 #include <linux/slab.h>
   4 #include <linux/blkdev.h>
   5 #include <linux/smp_lock.h>
   6 #include <linux/hdreg.h>
   7 #include <linux/virtio.h>
   8 #include <linux/virtio_blk.h>
   9 #include <linux/scatterlist.h>
  10
  11 #define PART_BITS 4
  12
  13 static int major, index;
  14
  15 struct virtio_blk
  16 {
  17         spinlock_t lock;
  18
  19         struct virtio_device *vdev;
  20         struct virtqueue *vq;
  21
  22         /* The disk structure for the kernel. */
  23         struct gendisk *disk;
  24
  25         /* Request tracking. */
  26         struct list_head reqs;
  27
  28         mempool_t *pool;
  29
  30         /* What host tells us, plus 2 for header & tailer. */
  31         unsigned int sg_elems;
  32
  33         /* Scatterlist: can be too big for stack. */
  34         struct scatterlist sg[/*sg_elems*/];
  35 };
  36
  37 struct virtblk_req
  38 {
  39         struct list_head list;
  40         struct request *req;
  41         struct virtio_blk_outhdr out_hdr;
  42         struct virtio_scsi_inhdr in_hdr;
  43         u8 status;
  44 };
  45
  46 static void blk_done(struct virtqueue *vq)
  47 {
  48         struct virtio_blk *vblk = vq->vdev->priv;
  49         struct virtblk_req *vbr;
  50         unsigned int len;
  51         unsigned long flags;
  52
  53         spin_lock_irqsave(&vblk->lock, flags);
  54         while ((vbr = virtqueue_get_buf(vblk->vq, &len)) != NULL) {
  55                 int error;
  56
  57                 switch (vbr->status) {
  58                 case VIRTIO_BLK_S_OK:
  59                         error = 0;
  60                         break;
  61                 case VIRTIO_BLK_S_UNSUPP:
  62                         error = -ENOTTY;
  63                         break;
  64                 default:
  65                         error = -EIO;
  66                         break;
  67                 }
  68
  69                 switch (vbr->req->cmd_type) {
  70                 case REQ_TYPE_BLOCK_PC:
  71                         vbr->req->resid_len = vbr->in_hdr.residual;
  72                         vbr->req->sense_len = vbr->in_hdr.sense_len;
  73                         vbr->req->errors = vbr->in_hdr.errors;
  74                         break;
  75                 case REQ_TYPE_SPECIAL:
  76                         vbr->req->errors = (error != 0);
  77                         break;
  78                 default:
  79                         break;
  80                 }
  81
  82                 __blk_end_request_all(vbr->req, error);
  83                 list_del(&vbr->list);
  84                 mempool_free(vbr, vblk->pool);
  85         }
  86         /* In case queue is stopped waiting for more buffers. */
  87         blk_start_queue(vblk->disk->queue);
  88         spin_unlock_irqrestore(&vblk->lock, flags);
  89 }
  90
  91 static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
  92                    struct request *req)
  93 {
  94         unsigned long num, out = 0, in = 0;
  95         struct virtblk_req *vbr;
  96
  97         vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
  98         if (!vbr)
  99                 /* When another request finishes we'll try again. */
 100                 return false;
 101
 102         vbr->req = req;
 103
 104         if (req->cmd_flags & REQ_FLUSH) {
 105                 vbr->out_hdr.type = VIRTIO_BLK_T_FLUSH;
 106                 vbr->out_hdr.sector = 0;
 107                 vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 108         } else {
 109                 switch (req->cmd_type) {
 110                 case REQ_TYPE_FS:
 111                         vbr->out_hdr.type = 0;
 112                         vbr->out_hdr.sector = blk_rq_pos(vbr->req);
 113                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 114                         break;
 115                 case REQ_TYPE_BLOCK_PC:
 116                         vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
 117                         vbr->out_hdr.sector = 0;
 118                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 119                         break;
 120                 case REQ_TYPE_SPECIAL:
 121                         vbr->out_hdr.type = VIRTIO_BLK_T_GET_ID;
 122                         vbr->out_hdr.sector = 0;
 123                         vbr->out_hdr.ioprio = req_get_ioprio(vbr->req);
 124                         break;
 125                 default:
 126                         /* We don't put anything else in the queue. */
 127                         BUG();
 128                 }
 129         }
 130
 131         if (vbr->req->cmd_flags & REQ_HARDBARRIER)
 132                 vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
 133
 134         sg_set_buf(&vblk->sg[out++], &vbr->out_hdr, sizeof(vbr->out_hdr));
 135
 136         /*
 137          * If this is a packet command we need a couple of additional headers.
 138          * Behind the normal outhdr we put a segment with the scsi command
 139          * block, and before the normal inhdr we put the sense data and the
 140          * inhdr with additional status information before the normal inhdr.
 141          */
 142         if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC)
 143                 sg_set_buf(&vblk->sg[out++], vbr->req->cmd, vbr->req->cmd_len);
 144
 145         num = blk_rq_map_sg(q, vbr->req, vblk->sg + out);
 146
 147         if (vbr->req->cmd_type == REQ_TYPE_BLOCK_PC) {
 148                 sg_set_buf(&vblk->sg[num + out + in++], vbr->req->sense, 96);
 149                 sg_set_buf(&vblk->sg[num + out + in++], &vbr->in_hdr,
 150                            sizeof(vbr->in_hdr));
 151         }
 152
 153         sg_set_buf(&vblk->sg[num + out + in++], &vbr->status,
 154                    sizeof(vbr->status));
 155
 156         if (num) {
 157                 if (rq_data_dir(vbr->req) == WRITE) {
 158                         vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
 159                         out += num;
 160                 } else {
 161                         vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
 162                         in += num;
 163                 }
 164         }
 165
 166         if (virtqueue_add_buf(vblk->vq, vblk->sg, out, in, vbr) < 0) {
 167                 mempool_free(vbr, vblk->pool);
 168                 return false;
 169         }
 170
 171         list_add_tail(&vbr->list, &vblk->reqs);
 172         return true;
 173 }
 174
 175 static void do_virtblk_request(struct request_queue *q)
 176 {
 177         struct virtio_blk *vblk = q->queuedata;
 178         struct request *req;
 179         unsigned int issued = 0;
 180
 181         while ((req = blk_peek_request(q)) != NULL) {
 182                 BUG_ON(req->nr_phys_segments + 2 > vblk->sg_elems);
 183
 184                 /* If this request fails, stop queue and wait for something to
 185                    finish to restart it. */
 186                 if (!do_req(q, vblk, req)) {
 187                         blk_stop_queue(q);
 188                         break;
 189                 }
 190                 blk_start_request(req);
 191                 issued++;
 192         }
 193
 194         if (issued)
 195                 virtqueue_kick(vblk->vq);
 196 }
 197
 198 /* return id (s/n) string for *disk to *id_str
 199  */
 200 static int virtblk_get_id(struct gendisk *disk, char *id_str)
 201 {
 202         struct virtio_blk *vblk = disk->private_data;
 203         struct request *req;
 204         struct bio *bio;
 205         int err;
 206
 207         bio = bio_map_kern(vblk->disk->queue, id_str, VIRTIO_BLK_ID_BYTES,
 208                            GFP_KERNEL);
 209         if (IS_ERR(bio))
 210                 return PTR_ERR(bio);
 211
 212         req = blk_make_request(vblk->disk->queue, bio, GFP_KERNEL);
 213         if (IS_ERR(req)) {
 214                 bio_put(bio);
 215                 return PTR_ERR(req);
 216         }
 217
 218         req->cmd_type = REQ_TYPE_SPECIAL;
 219         err = blk_execute_rq(vblk->disk->queue, vblk->disk, req, false);
 220         blk_put_request(req);
 221
 222         return err;
 223 }
 224
 225 static int virtblk_locked_ioctl(struct block_device *bdev, fmode_t mode,
 226                          unsigned cmd, unsigned long data)
 227 {
 228         struct gendisk *disk = bdev->bd_disk;
 229         struct virtio_blk *vblk = disk->private_data;
 230
 231         /*
 232          * Only allow the generic SCSI ioctls if the host can support it.
 233          */
 234         if (!virtio_has_feature(vblk->vdev, VIRTIO_BLK_F_SCSI))
 235                 return -ENOTTY;
 236
 237         return scsi_cmd_ioctl(disk->queue, disk, mode, cmd,
 238                               (void __user *)data);
 239 }
 240
 241 static int virtblk_ioctl(struct block_device *bdev, fmode_t mode,
 242                              unsigned int cmd, unsigned long param)
 243 {
 244         int ret;
 245
 246         lock_kernel();
 247         ret = virtblk_locked_ioctl(bdev, mode, cmd, param);
 248         unlock_kernel();
 249
 250         return ret;
 251 }
 252
 253 /* We provide getgeo only to please some old bootloader/partitioning tools */
 254 static int virtblk_getgeo(struct block_device *bd, struct hd_geometry *geo)
 255 {
 256         struct virtio_blk *vblk = bd->bd_disk->private_data;
 257         struct virtio_blk_geometry vgeo;
 258         int err;
 259
 260         /* see if the host passed in geometry config */
 261         err = virtio_config_val(vblk->vdev, VIRTIO_BLK_F_GEOMETRY,
 262                                 offsetof(struct virtio_blk_config, geometry),
 263                                 &vgeo);
 264
 265         if (!err) {
 266                 geo->heads = vgeo.heads;
 267                 geo->sectors = vgeo.sectors;
 268                 geo->cylinders = vgeo.cylinders;
 269         } else {
 270                 /* some standard values, similar to sd */
 271                 geo->heads = 1 << 6;
 272                 geo->sectors = 1 << 5;
 273                 geo->cylinders = get_capacity(bd->bd_disk) >> 11;
 274         }
 275         return 0;
 276 }
 277
 278 static const struct block_device_operations virtblk_fops = {
 279         .ioctl  = virtblk_ioctl,
 280         .owner  = THIS_MODULE,
 281         .getgeo = virtblk_getgeo,
 282 };
 283
 284 static int index_to_minor(int index)
 285 {
 286         return index << PART_BITS;
 287 }
 288
 289 static ssize_t virtblk_serial_show(struct device *dev,
 290                                 struct device_attribute *attr, char *buf)
 291 {
 292         struct gendisk *disk = dev_to_disk(dev);
 293         int err;
 294
 295         /* sysfs gives us a PAGE_SIZE buffer */
 296         BUILD_BUG_ON(PAGE_SIZE < VIRTIO_BLK_ID_BYTES);
 297
 298         buf[VIRTIO_BLK_ID_BYTES] = '\0';
 299         err = virtblk_get_id(disk, buf);
 300         if (!err)
 301                 return strlen(buf);
 302
 303         if (err == -EIO) /* Unsupported? Make it empty. */
 304                 return 0;
 305
 306         return err;
 307 }
 308 DEVICE_ATTR(serial, S_IRUGO, virtblk_serial_show, NULL);
 309
 310 static int __devinit virtblk_probe(struct virtio_device *vdev)
 311 {
 312         struct virtio_blk *vblk;
 313         struct request_queue *q;
 314         int err;
 315         u64 cap;
 316         u32 v, blk_size, sg_elems, opt_io_size;
 317         u16 min_io_size;
 318         u8 physical_block_exp, alignment_offset;
 319
 320         if (index_to_minor(index) >= 1 << MINORBITS)
 321                 return -ENOSPC;
 322
 323         /* We need to know how many segments before we allocate. */
 324         err = virtio_config_val(vdev, VIRTIO_BLK_F_SEG_MAX,
 325                                 offsetof(struct virtio_blk_config, seg_max),
 326                                 &sg_elems);
 327
 328         /* We need at least one SG element, whatever they say. */
 329         if (err || !sg_elems)
 330                 sg_elems = 1;
 331
 332         /* We need an extra sg elements at head and tail. */
 333         sg_elems += 2;
 334         vdev->priv = vblk = kmalloc(sizeof(*vblk) +
 335                                     sizeof(vblk->sg[0]) * sg_elems, GFP_KERNEL);
 336         if (!vblk) {
 337                 err = -ENOMEM;
 338                 goto out;
 339         }
 340
 341         INIT_LIST_HEAD(&vblk->reqs);
 342         spin_lock_init(&vblk->lock);
 343         vblk->vdev = vdev;
 344         vblk->sg_elems = sg_elems;
 345         sg_init_table(vblk->sg, vblk->sg_elems);
 346
 347         /* We expect one virtqueue, for output. */
 348         vblk->vq = virtio_find_single_vq(vdev, blk_done, "requests");
 349         if (IS_ERR(vblk->vq)) {
 350                 err = PTR_ERR(vblk->vq);
 351                 goto out_free_vblk;
 352         }
 353
 354         vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
 355         if (!vblk->pool) {
 356                 err = -ENOMEM;
 357                 goto out_free_vq;
 358         }
 359
 360         /* FIXME: How many partitions?  How long is a piece of string? */
 361         vblk->disk = alloc_disk(1 << PART_BITS);
 362         if (!vblk->disk) {
 363                 err = -ENOMEM;
 364                 goto out_mempool;
 365         }
 366
 367         q = vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
 368         if (!q) {
 369                 err = -ENOMEM;
 370                 goto out_put_disk;
 371         }
 372
 373         q->queuedata = vblk;
 374
 375         if (index < 26) {
 376                 sprintf(vblk->disk->disk_name, "vd%c", 'a' + index % 26);
 377         } else if (index < (26 + 1) * 26) {
 378                 sprintf(vblk->disk->disk_name, "vd%c%c",
 379                         'a' + index / 26 - 1, 'a' + index % 26);
 380         } else {
 381                 const unsigned int m1 = (index / 26 - 1) / 26 - 1;
 382                 const unsigned int m2 = (index / 26 - 1) % 26;
 383                 const unsigned int m3 =  index % 26;
 384                 sprintf(vblk->disk->disk_name, "vd%c%c%c",
 385                         'a' + m1, 'a' + m2, 'a' + m3);
 386         }
 387
 388         vblk->disk->major = major;
 389         vblk->disk->first_minor = index_to_minor(index);
 390         vblk->disk->private_data = vblk;
 391         vblk->disk->fops = &virtblk_fops;
 392         vblk->disk->driverfs_dev = &vdev->dev;
 393         index++;
 394
 395         if (virtio_has_feature(vdev, VIRTIO_BLK_F_FLUSH)) {
 396                 /*
 397                  * If the FLUSH feature is supported we do have support for
 398                  * flushing a volatile write cache on the host.  Use that
 399                  * to implement write barrier support.
 400                  */
 401                 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN_FLUSH);
 402         } else if (virtio_has_feature(vdev, VIRTIO_BLK_F_BARRIER)) {
 403                 /*
 404                  * If the BARRIER feature is supported the host expects us
 405                  * to order request by tags.  This implies there is not
 406                  * volatile write cache on the host, and that the host
 407                  * never re-orders outstanding I/O.  This feature is not
 408                  * useful for real life scenarious and deprecated.
 409                  */
 410                 blk_queue_ordered(q, QUEUE_ORDERED_TAG);
 411         } else {
 412                 /*
 413                  * If the FLUSH feature is not supported we must assume that
 414                  * the host does not perform any kind of volatile write
 415                  * caching. We still need to drain the queue to provider
 416                  * proper barrier semantics.
 417                  */
 418                 blk_queue_ordered(q, QUEUE_ORDERED_DRAIN);
 419         }
 420
 421         /* If disk is read-only in the host, the guest should obey */
 422         if (virtio_has_feature(vdev, VIRTIO_BLK_F_RO))
 423                 set_disk_ro(vblk->disk, 1);
 424
 425         /* Host must always specify the capacity. */
 426         vdev->config->get(vdev, offsetof(struct virtio_blk_config, capacity),
 427                           &cap, sizeof(cap));
 428
 429         /* If capacity is too big, truncate with warning. */
 430         if ((sector_t)cap != cap) {
 431                 dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
 432                          (unsigned long long)cap);
 433                 cap = (sector_t)-1;
 434         }
 435         set_capacity(vblk->disk, cap);
 436
 437         /* We can handle whatever the host told us to handle. */
 438         blk_queue_max_segments(q, vblk->sg_elems-2);
 439
 440         /* No need to bounce any requests */
 441         blk_queue_bounce_limit(q, BLK_BOUNCE_ANY);
 442
 443         /* No real sector limit. */
 444         blk_queue_max_hw_sectors(q, -1U);
 445
 446         /* Host can optionally specify maximum segment size and number of
 447          * segments. */
 448         err = virtio_config_val(vdev, VIRTIO_BLK_F_SIZE_MAX,
 449                                 offsetof(struct virtio_blk_config, size_max),
 450                                 &v);
 451         if (!err)
 452                 blk_queue_max_segment_size(q, v);
 453         else
 454                 blk_queue_max_segment_size(q, -1U);
 455
 456         /* Host can optionally specify the block size of the device */
 457         err = virtio_config_val(vdev, VIRTIO_BLK_F_BLK_SIZE,
 458                                 offsetof(struct virtio_blk_config, blk_size),
 459                                 &blk_size);
 460         if (!err)
 461                 blk_queue_logical_block_size(q, blk_size);
 462         else
 463                 blk_size = queue_logical_block_size(q);
 464
 465         /* Use topology information if available */
 466         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 467                         offsetof(struct virtio_blk_config, physical_block_exp),
 468                         &physical_block_exp);
 469         if (!err && physical_block_exp)
 470                 blk_queue_physical_block_size(q,
 471                                 blk_size * (1 << physical_block_exp));
 472
 473         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 474                         offsetof(struct virtio_blk_config, alignment_offset),
 475                         &alignment_offset);
 476         if (!err && alignment_offset)
 477                 blk_queue_alignment_offset(q, blk_size * alignment_offset);
 478
 479         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 480                         offsetof(struct virtio_blk_config, min_io_size),
 481                         &min_io_size);
 482         if (!err && min_io_size)
 483                 blk_queue_io_min(q, blk_size * min_io_size);
 484
 485         err = virtio_config_val(vdev, VIRTIO_BLK_F_TOPOLOGY,
 486                         offsetof(struct virtio_blk_config, opt_io_size),
 487                         &opt_io_size);
 488         if (!err && opt_io_size)
 489                 blk_queue_io_opt(q, blk_size * opt_io_size);
 490
 491
 492         add_disk(vblk->disk);
 493         err = device_create_file(disk_to_dev(vblk->disk), &dev_attr_serial);
 494         if (err)
 495                 goto out_del_disk;
 496
 497         return 0;
 498
 499 out_del_disk:
 500         del_gendisk(vblk->disk);
 501         blk_cleanup_queue(vblk->disk->queue);
 502 out_put_disk:
 503         put_disk(vblk->disk);
 504 out_mempool:
 505         mempool_destroy(vblk->pool);
 506 out_free_vq:
 507         vdev->config->del_vqs(vdev);
 508 out_free_vblk:
 509         kfree(vblk);
 510 out:
 511         return err;
 512 }
 513
 514 static void __devexit virtblk_remove(struct virtio_device *vdev)
 515 {
 516         struct virtio_blk *vblk = vdev->priv;
 517
 518         /* Nothing should be pending. */
 519         BUG_ON(!list_empty(&vblk->reqs));
 520
 521         /* Stop all the virtqueues. */
 522         vdev->config->reset(vdev);
 523
 524         del_gendisk(vblk->disk);
 525         blk_cleanup_queue(vblk->disk->queue);
 526         put_disk(vblk->disk);
 527         mempool_destroy(vblk->pool);
 528         vdev->config->del_vqs(vdev);
 529         kfree(vblk);
 530 }
 531
 532 static const struct virtio_device_id id_table[] = {
 533         { VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
 534         { 0 },
 535 };
 536
 537 static unsigned int features[] = {
 538         VIRTIO_BLK_F_BARRIER, VIRTIO_BLK_F_SEG_MAX, VIRTIO_BLK_F_SIZE_MAX,
 539         VIRTIO_BLK_F_GEOMETRY, VIRTIO_BLK_F_RO, VIRTIO_BLK_F_BLK_SIZE,
 540         VIRTIO_BLK_F_SCSI, VIRTIO_BLK_F_FLUSH, VIRTIO_BLK_F_TOPOLOGY
 541 };
 542
 543 /*
 544  * virtio_blk causes spurious section mismatch warning by
 545  * simultaneously referring to a __devinit and a __devexit function.
 546  * Use __refdata to avoid this warning.
 547  */
 548 static struct virtio_driver __refdata virtio_blk = {
 549         .feature_table = features,
 550         .feature_table_size = ARRAY_SIZE(features),
 551         .driver.name =  KBUILD_MODNAME,
 552         .driver.owner = THIS_MODULE,
 553         .id_table =     id_table,
 554         .probe =        virtblk_probe,
 555         .remove =       __devexit_p(virtblk_remove),
 556 };
 557
 558 static int __init init(void)
 559 {
 560         major = register_blkdev(0, "virtblk");
 561         if (major < 0)
 562                 return major;
 563         return register_virtio_driver(&virtio_blk);
 564 }
 565
 566 static void __exit fini(void)
 567 {
 568         unregister_blkdev(major, "virtblk");
 569         unregister_virtio_driver(&virtio_blk);
 570 }
 571 module_init(init);
 572 module_exit(fini);
 573
 574 MODULE_DEVICE_TABLE(virtio, id_table);
 575 MODULE_DESCRIPTION("Virtio block driver");
 576 MODULE_LICENSE("GPL");